當我們分析一筆資料時，常常會根據資料的型態，對資料作參數模型的假設，但是當模型假設錯誤時，分析的結果會有錯誤。本文在Royall and Tsou (2003)的強韌概似函數理念下，針對不同類型的非負連續資料，以離散型分配卜瓦松與負二項的強韌概似函數提供正確的統計推論。經由模擬發現，這兩類模型表現比連續型分配的常態強韌概似函數好，而在一些情況下，表現不比連續型伽瑪強韌概似函數與逆高斯強韌概似函數差。文中也以不適合強韌化的連續型對數常態分配作為反例，說明可強韌化條件(Royall and Tsou, 2003)的重要。 The purpose of this research is trying to use discrete distribution to construct a likelihood function for non-negative continuous data. We focus on the Poisson distribution and negative binomial distribution and use the robust likelihood methodology introduced by Royall and Tsou (2003). Finally, we can see that the robust Poisson model and the robust negative binomial model are more efficient than the robust normal model. Moreover, we use a counter-example to illustrate that it is not coincidental.
